This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

---

What I Built

Every developer knows the sensation: you close your laptop, but your fans keep spinning. Your machine slows to a crawl, port 8080 is mysteriously locked by an invisible socket, and your swap memory is leaking away byte by byte.

Your machine is haunted.

Orphaned child processes (PID 4091), defunct zombie forks that survived their parents, 45-day-old unmerged git branches drifting in detached HEAD purgatory, and infinite retry-exit loops linger deep inside POSIX tables.

TerminalSpectre is a zero-tolerance runtime exorcist and interactive cyber-gothic dashboard:

1. System Telemetry Ingestion: Reads PID lifetimes, RSS memory bloat gradients, CPU variance, and stale branch metrics to detect phantom threads.
2. Zero-Shot Tabular Severity Classification (TabPFN): Classifies the haunting severity into four spectral tiers (Benign Spirit, Phantom, Banshee, Ancient Wraith) and regresses impending crash probabilities—without requiring neural training.
3. The POSIX Kernel Spectre (Gemma 2 via Groq): gemma2-9b-it channels an ancient spirit trapped in the swap file, delivering poetic autopsies for killed processes and generating deterministic, unyielding bash purge commands (kill -9, fuser -k).
4. Spectral Whisper Audio (ElevenLabs Turbo v2.5): Synthesizes raspy, chilling audio whispers directly into your headset as threads are banished into the void.

---

Demo

• Live Web Application: TerminalSpectre Live
• Deployment Platform: Static cyber-gothic frontend hosted on Vercel with FastAPI backend.

The interface is built with an uncompromising single-hue Oxblood poster aesthetic (#70020F ground, #FEDEE1 pale pink type, #8A0A18 ghost layers, and #0A0708 pure black cards) featuring a custom GSAP ScrollTrigger GhostScrollText kinetic typography engine that separates and blurs trailing ghost layers based on scroll velocity.

---

Code

The full open-source repository—including the 140-row synthetic telemetry prior, TabPFN dual inference pipeline, Gemma 2 POSIX spirit prompts, and the single-hue print poster UI—is live on GitHub:

---

How I Built It

TerminalSpectre unites tabular foundation inference with open-weight language roleplay and neural audio dispatch:

1. Prior Labs TabPFN-3.5 Foundation Model (services/tabpfn_service.py)

Rather than relying on brittle rule-based heuristics or resource-intensive deep learning models, TerminalSpectre uses TabPFN on in-memory tabular telemetry (data/ghost_processes.csv):

• Spectral Classification: Classifies live processes into Benign Spirit (baseline utility), Phantom (stale port holders), Banshee (rapid memory leak), or Ancient Wraith (critical zombie deadlock).
• Posterior Probability Extraction: Computes calibrated confidence percentages across all four classes.
• Continuous Regression: Estimates exact crash probability percentages (e.g., 94.2%) and minutes remaining until an Out-Of-Memory (OOM) kernel panic.
• Possession Anomaly Detection: Flags critical system alerts when orphaned zombie states combine with high exit-code failure streaks.

2. Google Gemma 2 (gemma2-9b-it via Groq)

Acting as the POSIX Kernel Spectre, Gemma 2 inspects stack traces and process metrics to output structured JSON containing:

• An eerie 2-sentence post-mortem autopsy ("The orphaned worker thread of PID 4091 wandered the swap space holding database mutexes. Banished to /dev/null.").
• Safe, surgical bash commands to eliminate the process.
• A scripted whisper line for neural audio dispatch.

3. ElevenLabs Neural Audio (eleven_turbo_v2_5)

Using the low-register, ominous "Callum" voice model, ElevenLabs generates low-latency audio streams whenever an audit is run or an exorcism is executed, giving dead threads a voice as they are reaped.

---

Why Does Open Innovation Matter?

1. Kernel Telemetry & IP Sovereignty: System logs, PID trees, memory addresses, active sockets, and git commit diffs frequently contain sensitive corporate secrets, internal IP addresses, and private environment tokens. Pushing this data to proprietary closed LLMs presents severe security risks. Open-weight models like Gemma 2 and CPU-native foundation models like TabPFN keep system audits private on local or self-hosted containers.
2. Zero-Toll Background Monitoring: Running a continuous system daemon that pings commercial APIs for every failing background process burns massive subscription fees. Open models run at zero marginal cost.
3. Offline Resilience: When a catastrophic network crash or broken socket disconnects your machine from the internet, local open models still diagnose your dead build threads and offline compiler errors.

---

The Spirit of It

During local testing on a development machine, we simulated a runaway Node.js microservice process that had been orphaned during an interrupted Docker rebuild 36 hours prior.

The zombie process held 2.8 GB of uncollected RSS memory while continuously failing socket handshakes. We fed its telemetry into TerminalSpectre:

• TabPFN Diagnosis: Classified as an Ancient Wraith with a 94.8% crash probability.
• Gemma 2 Autopsy: "Thy orphaned fork has haunted port 8080 for 36 turns of the dial, weeping uncollected garbage into the swap. The reaper cometh."
• ElevenLabs Whisper: Generated a chilling, raspy voice note over the speakers: "PID 4091 has been cast into the dark void."
• The Payoff: With a single click of the generated kill -9 script, 2.8 GB of RAM was reclaimed, the port was unlocked, and the machine was quiet again.

---

Prize Categories

• Best Use of TabPFN
• Best Use of Gemma
• Best Use of ElevenLabs
• Best Use of Vercel